dissenting:
I respectfully dissent.
In In re Sumitomo Copper Litigation, we recently held that “petitioners seeking leave to appeal pursuant to Rule 23(f) must demonstrate either (1) that the certification order will effectively terminate the litigation and there has been a substantial showing that the district court’s decisions is questionable, or (2) that the certification order implicates a legal question about which there is a compelling need for immediate resolution.” 262 F.3d at 139, (2d Cir.2001) This approach accords with the advisory committee’s note to the Rule: “Permission to appeal ... is most likely to be granted when the certification decision turns on a novel or unsettled question of law, or when, as a practical matter, the decision on certification is likely dispositive of the litigation.” Fed.R.Civ.P. 23 advisory committee’s note.
Appellants here have made a more than sufficient showing under Sumitomo (A) that the certification order tends to termi*148nate the litigation by a coercive settlement and (B) that two crucial rulings (embedded in the district court’s otherwise thorough opinion) are questionable, to say the least.
(A) Rule 23(f) was adopted in part to alleviate the danger that an erroneous “order granting certification” may force a defendant “to settle rather than incur the costs of defending a class action and run the risk of ruinous liability.” Fed.R.Civ.P. 23 advisory committee’s note. One sound basis for granting jurisdiction under Rule 23(f) is thus the circumstance that the class certification “places inordinate or hydraulic pressure on defendants to settle, avoiding the risk, however small, of potentially ruinous liability.” Newton v. Merrill Lynch, 259 F.3d 154, 164 (3rd Cir.2001). No doubt, true “death knell” cases are few; and a reviewing court “must be wary lest the mind hear a bell that is not tolling.” Blair v. Equifax Check Servs., Inc., 181 F.3d 832, 834-35 (7th Cir.1999). But in this case-in which the aggregated and trebled claims of the four million class members are alleged to top $100 billion — a carillon is in full peal. Even a defendant who is innocent and holy may rationally choose to pay a few hundred million dollars in settlement of a class action rather than “run the risk of ruinous liability.” As the district court observed (without the slightest overstatement), the “enormous financial risks” faced by the defendants “are obviously increased drastically by certification of the class.” In re Visa Check/MasterMoney Antitrust Litig., 192 F.R.D. at 89. The certification order in this case may therefore coerce settlement.
(B) The district court’s decision is rendered “questionable” by two errors of law that render grant of the certification an abuse of discretion.
First, the district court inadequately considered whether the class action can be manageably tried and has certified the four million member class without identifying a “ ‘practical means’ ” for calculating the extent of each merchant’s damages. See Abrams v. Interco, 719 F.2d 23, 31 (2d Cir.1983) (Friendly, J.) (quoting II Areeda & Turner, Antitrust Law ¶3320, at 157 (1978)). The majority opinion draws comfort from the conditional nature of the certification and the district court’s retained power to modify it. But this is cold comfort for the defendants: the brutally coercive effect of the certification cannot be stayed.
Second, the district court ruled on certification without deciding a threshold question of law that determines whether the class members’ interests will be aligned and therefore whether litigation by representative is appropriate. Courts are split on whether a tying plaintiffs injury must be measured (as I believe) by how much the tie increased the combined cost of buying both the tying and the tied products. If that is the correct measure, there can be no class here because the putative members used the tying and tied products in different proportions, and therefore have conflicting interests when it comes to how the fact and extent of any injuries are to be proven.
Because of these two legal errors, and because the majority opinion is so deferential that this appeal is no check at all on the coercion that Rule 23(f) is designed to curb, I cannot join the majority opinion.
I.
Certification Prior to Identification of a “Practical Means” for Trying Four Million Claims
In ruling on certification, a court is required to consider “the difficulties likely to be encountered in the management of a class action.” Fed. R. Civ. Pro. 23(b)(3)(D). And unless a court is “con*149fronted with a request for a settlement-only class certification,” the court must specifically “inquire whether the case, if tried, would present intractable management problems.” Amchem Prods, v. Windsor, 521 U.S. 591, 620, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997) (emphasis added) (noting that the inquiry is unnecessary as to “settlement-only” classes, because “the proposal is that there be no trial” (citing Fed. R. Civ. Pro. 23(b)(3)(D))). For that reason, Judge Friendly instructed that if an antitrust “‘class numbers in the thousands or millions,’ ” a court may not grant certification until “ ‘it can devise a practical means’ ” for trying the action.1 Abrams, 719 F.2d at 31 (quoting II Areeda & Turner, supra ¶ 332c, at 157).
Any case of this size and numerosity is beset by potentially intractable problems of manageability. The crucial, unresolved difficulty in this litigation concerns individualized proof of damages, an issue that “can be critical in determining whether a class of antitrust claimants should be certified.” Manual for Complex Litigation § 33.11, at 303 (3d ed.1995); see Windham v. American Brands, Inc., 565 F.2d 59, 71 (4th Cir.1977) (en banc) (holding that “in determining the manageability of a proposed [antitrust] class action,” the district court must “consider proof of damages”). Specifically, a defense has been interposed that if litigated would require findings as to the extent of the damages recoverable by each of the four million members of the putative class.2
The law is “settled that affirmative defenses should be considered in making class certification decisions.” Waste Mgmt. Holdings v. Mowbray, 208 F.3d 288, 295 (1st Cir.2000); see also Majority Op. at 138, 139 (a court must examine “defenses in determining whether a putative class meets the requirements of Rule 23(b)(3)”) (emphasis added); Castano v. American Tobacco Co., 84 F.3d 734, 744 (5th Cir.1996) (“Going beyond the pleadings is necessary, as a court must understand [inter alia] defenses ... in order to make a meaningful determination of the certification issues.”); cf. Broussard v. Meineke Discount Muffler Shops, 155 F.3d 331, 342 (4th Cir.1998) (certification is “erroneous” when affirmative defenses “may depend on facts peculiar to each plaintiffs case”) (citation omitted). And there is good reason for considering the manageability of defenses in this case: even if each merchant’s claim took no more than a half a day to sort out, the damages phase of trial would last as long as the whole course of Western civilization from Ur.
The asserted defense — which the district judge acknowledged may “have force” — is that the merchants had a duty to mitigate their losses by encouraging at least some *150of their customers to use forms of payment other than off-line debit. In re Visa Check/Mastermoney Antitrust Litig., 192 F.R.D. at 86. We have repeatedly held that mitigation is a viable theory in antitrust actions. See Litton Systems v. AT & T, 700 F.2d 785, 820 n. 47 (2d Cir.1983) (Sherman Act § 2); Borger v. Yamaha Int’l Corp., 625 F.2d 390, 399 (2d Cir.1980) (Sherman Act § 1); see generally Neil W. Hamilton & Virginia B. Cone, Mitigation of Antitrust Damages, 66 Or.L.Rev. 339 (1984). Enforcement of the mitigation duty may be even more critical in antitrust cases than in cases sounding in contract or tort because an antitrust plaintiff seeking treble damages can profit by avoiding mitigation of loss. See Hamilton & Cone, supra at 355 (noting “the perverse impact of trebling on a plaintiffs incentive to minimize damage”).
Just as we assume for purposes of class certification that the illegality of the tie is litigable, we must assume that defendants will be entitled to try their mitigation defense. The rule against pre-certification decisions on the merits is non-partisan and serves neutrally to avoid the risk that a “court’s tentative findings, made in the absence of established safeguards, may color the subsequent proceedings.” Eisen v. Carlisle & Jacquelin, 417 U.S. 156, 178, 94 S.Ct. 2140, 40 L.Ed.2d 732 (1974).
The majority recites the principle that a court “must examine ... defenses in determining whether a putative class meets the requirements of Rule 23(b)(3),” but finds no abuse of discretion because the district court “specifically recognized its ability to modify its class certification order, sever liability and damages, or even decertify the class if such an action ultimately became necessary.” Majority Op. at 138, 139 (emphasis added) (citing In re Visa Check/MasterMoney Antitrust Litig., 192 F.R.D. at 89 (citing Fed. R. Civ. Pro. 23(c)(1))).
This approach reflects a serious misinterpretation of Rule 23(c)(1), which permits courts to enter a “conditional” certification and then alter or amend the certification as the case proceeds. But as other circuits have repeatedly explained, individual questions “must be resolved before a class is certified, even if certification is conditional.” In re Hotel Tel. Charges, 500 F.2d 86, 90 (9th Cir.1974) (emphasis added); accord Windham v. American Brands, Inc., 565 F.2d 59, 70 (4th Cir.1977) (en banc). Conditional certification therefore is emphatically not an expedient by which a court may “ ‘avoid deciding whether, at that time, the requirements of the Rule have been substantially met.’ ” Castano v. American Tobacco Co., 84 F.3d 734, 741 (5th Cir.1996) (quoting In re Hotel Tel. Charges, 500 F.2d at 90); accord In re American Med. Sys., 75 F.3d 1069, 1088 n. 21 (6th Cir.1996); see Manual for Complex Litigation § 30.11, at 215 (“Although Fed.R.Civ.P. 23(c)(1) permits entry of a ‘conditional’ class determination order and amendment before the decision on the merits, that procedure should not be used to defer a final class ruling.”) (emphasis added). “ ‘The purpose of conditional certification’ ” is, rather, “ ‘to preserve the Court’s power to revoke certification in those cases wherein the magnitude or complexity of the litigation may eventually reveal problems not theretofore apparent.’ ” Castano, 84 F.3d at 741 (emphasis added) (quoting in re Hotel Tel. Charges, 500 F.2d at 90).
The manageability problem in this case has been “apparent” from the outset and is apparent right now. The record developed so far reveals that Mastercard encouraged merchants to channel customers *151to the company’s on-line debit alternative, which it explained was “more secure and therefore less expensive.” And as plaintiffs conceded in the district court, some merchants did adopt programs that so channeled an appreciable number of would-be off-line debit transactions, thus reducing the amount of damages that those class members could collect from this suit. See Plaintiffs’ Reply Memorandum of Law in Support in of Plaintiffs’ Motion for Class Certification at 36-37 & n. 40. The mitigation issues are whether it is reasonable to expect other class members to have adopted similar programs, and to what extent would such programs have achieved mitigation for such class members. The district judge’s recognition that the defense may “have force” is well-founded.3
It makes no sense to provide individualized notice to millions of merchants before the court has devised a “practical means” for trying the issues presented. See Fed. R. Civ. Pro. 23(c)(2) (requiring “individual notice to all [Rule 23(b)(3) damages class] members who can be identified through reasonable effort”). Certification that is premature does no favor to the millions of absent class members because “[undesirable consequences may follow” if the class “is later decertified or redefined.” Manual for Complex Litigation § 30.11, at 215. Class members who are “eliminated from the litigation as a result of decertification or reduction in the size of a class may be confused at best or prejudiced at worst.” Id. Even if relief is eventually obtained for a pared down class, “those who were initially in the larger class may attempt to reverse the decision that excluded them from the class; such a reversal may be particularly troublesome if the relief was obtained by settlement.” Id.
The majority opinion’s view of conditional certification also seriously impairs our ability to conduct a meaningful interlocutory review under Rule 23(f). Conditional certification is not a hedge against present error. Our power of interlocutory review was conferred for a case such as this, where scores of billions of dollars are thought to be at stake and no procedure has been conceived for a fair merits adjudication. A purpose of our interlocutory look is to assure that defendants are not coerced into settlement by an improvident certification, conditional or otherwise. But if district courts can simply postpone consideration of difficult issues by making certification “conditional” (as all certifications tend to be anyway), it is unlikely that we would ever vacate such an order. At a minimum, the approach in the majority opinion will lead unnecessarily to multiple requests for appellate intervention in large-scale class actions: one after the conditional certification, and others as the district judge eventually confronts issues that were apparent before conditional certification was given.
Finally, unless there is a “practical means” of trying the case, certification leads to unfairness and injustice: if the district court later finds that a trial is unmanageable, the named plaintiffs and their counsel lose nothing, but in the meantime they have in hand the means to extract a favorable settlement of what may be weak claims. See Eisen v. Carlisle & Jacquelin, 417 U.S. 156, 177, 94 S.Ct. 2140, 40 L.Ed.2d 732 (1974) (explaining that a “representative plaintiff’ should not be al*152lowed to “secure the benefits of a class action without first satisfying the requirements for it”). Once a certification “order is issued, the parties can be expected to rely on it and ... engage in settlement discussions on the assumption that in the normal course of events it will not be altered except for good cause.” Manual for Complex Litigation § 30.17, at 223. “Judge Friendly, who was not given to hyperbole, called settlements induced by a small probability of an immense judgment in a class action ‘blackmail settlements.’ ” In re Rhone-Poulenc Rorer, 51 F.3d 1293, 1298 (7th Cir.1995) (Posner, J.) (quoting Henry J. Friendly, Federal Jurisdiction: A General View 120 (1973)); accord Rutstein v. Avis Rent-A-Car Sys., 211 F.3d 1228, 1241 n. 21 (11th Cir.2000) (referring to “the blackmail value of a class certification that can aid the plaintiffs in coercing the defendant into a settlement”). Unfortunately, the majority’s excessive deference facilitates such an injustice. Cf. Szabo v. Bridgeport Machines, 249 F.3d 672, 675 (7th Cir.2001) (holding that a “prime occasion for Rule 23(f)” appellate review is when class certification “puts a bet-your-company decision to [defendant’s] managers and may induce a substantial settlement even if the [plaintiffs’] position is weak”).
Finally, -de-certification would not doom this litigation; this is not a case in which the named plaintiffs’ claims are “far smaller than the costs of litigation.” Fed. R.Civ.P. 23(f) advisory committee’s note. For instance, if named plaintiff Wal Mart Stores processed even 10% of its sales with credit and off-line debit, and if it established an (average) one-half percent overcharge on those transactions, its claim would reach a quarter billion dollars (with trebling and before attorneys’ fees) for 1999 alone.
II.
Conflicts Among Class Members in Proving the Fact and Extent of Injury
Certification of this class is untenable for another, entirely independent reason: under the correct measure of injury in a tying ease, the class members face irreconcilable conflicts in proving the fact and extent of any injury caused by the tie. Rule 23(a) therefore prohibits litigation by representative because no representative can adequately protect the interests of the millions of absent class members. See General Tel. Co. v. Falcon, 457 U.S. 147, 156-57 & n. 13, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982) (Rule 23(a) tests “whether the named plaintiffs claim and the class claims are so interrelated that the interests of the class members will be fairly and adequately protected in their absence”).
There is little dispute that plaintiffs can use common proof to litigate the threshold question as to whether there is an illegal tie.4 Consequently, the “principal” issue for consideration by the district court in weighing certification was whether the plaintiffs also could prove that the allegedly illegal tie caused the class “injury in fact” the “gravamen of a tie-in action.” In re Visa Check/MasterMoney Antitrust Litig., 192 F.R.D. at 81 (“[E]ven an antitrust plaintiff who has successfully proven illegal conduct on the part of the defendant must also show that it was ‘injured in fact’ by the illegality.”); Kypta v. McDonald’s Corp., 671 F.2d 1282, 1285 (11th Cir.1982). The injury in fact element is satisfied by *153“proof of some damage flowing from the” defendant’s conduct, see Zenith Radio Corp. v. Hazeltine Research, 395 U.S. 100, 114 n. 9, 89 S.Ct. 1562, 23 L.Ed.2d 129 (1969); under the Rules Enabling Act, this requirement applies to class actions as well as to litigation by named parties only. See Alabama v. Blue Bird Body Co.,. 573 F.2d 309, 318 (5th Cir.1978); Bogosian v. Gulf Oil Corp., 561 F.2d 434, 454 (3d Cir.1977). Plaintiffs therefore concede that their class of four million merchants cannot be certified unless plaintiffs rely on a theory that posits at least some injury to every class member. However, the district court avoided or has delayed deciding how injury in fact would be measured. In this litigation, that is a vexed question.
In many tying cases, measuring the impact of the tie is fairly mechanical because the buyers have used the two products in fixed proportions. See, e.g., United States v. Microsoft Corp., 147 F.3d 935 (D.C.Cir.1998) (one computer operating system; one internet web browser); Town Sound and Custom Tops v. Chrysler Motors Corp., 959 F.2d 468 (3d Cir.1992) (en banc) (one car; one automobile sound system). But the question is troublesome here, and should have been considered in the context of class certification, because these plaintiffs have used the credit cards (thé tying product) and off-line debit cards (the tied product) in different proportions; fine jewelers, for example, would rarely be presented with an off-line debit card. See In re Visa Check/MasterMoney Antitrust Litig., 192 F.R.D. at 83. The varying proportions of use create intractable conflicts among the four million merchants agglomerated as a class by the district court.
In my view: (1) The court’s failure to decide how injury in fact would be -measured was an error of law and thus an abuse of discretion; (2) the correct measure of injury is the extent to which a merchant would have paid less for the “package” of the tying and tied products in the absence of the tie; and (3) plaintiffs who use tied and tying products in varying proportions have conflicting interests in how the class goes about proving the fact that they would have paid less for their particular “package,” as well as in showing the extent of any disparity. Since Rule 23(a) requires that members of a class “possess the same interest and suffer the same injury,” class certification is not an option in this case. Amchem Prods. v. Windsor, 521 U.S. 591, 625-26, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997).
A. The “Measure of Injury” Applied in the District Court
The need to show injury to all class members has been problematic for plaintiffs from the outset, and their theory of injury has mutated as class counsel has tried gamely to fix the problem. The Second Amended Class Action Complaint alleged that “absent the tie, virtually all members of the putative class would have refused the off-line debit cards and processed the same transactions using cash, checks, on-line debit, or other payment forms.” In re Visa Check/MasterMoney Antitrust Litig., 192 F.R.D. at 82 (emphasis added). In the course of briefing the class certification motion, however, defendants demonstrated that there could be no class-wide injury under that scenario because (inter alia) “even some named plaintiffs have testified that they would have continued to accept off-line debit cards at current prices even absent the tie.” Id.
Plaintiffs then sponsored an expert’s affidavit positing that all plaintiffs would have continued to accept off-line debit cards even absent the tie, but that the defendants would have charged a lower fee for them. See id. Specifically, the expert *154articulated the following theory of injury in fact: “in the ‘but-for,’ untied world,” Visa and Mastercard “would have cut their interchange fees in order to preserve universal acceptance.” In re Visa Check/MasterMoney Antitrust Litig., 192 F.R.D. at 82. “That move,” according to the expert, “would have led merchants to accept the cards once again.” Id. This theory thus posited “class-wide injury resulting from every single class member’s overpaying for off-line debit cards as a direct result of the tie.” Id. Plaintiffs supported their argument by pointing to tying cases that measure injury solely by the inflated price paid for the tied product. See, e.g., Bell v. Cherokee Aviation Corp., 660 F.2d 1123, 1133 (6th Cir.1981) (“[T]he measure of damages is the difference between the price actually paid for the tied product and the price at which the product could have been obtained on the open market.”). But see X Phillip E. Areeda et al., Antitrust Law ¶ 1769c, at 430 (1996) (calling the “tied product” measure of injury “quite wrong”).
Defendants responded that the tied product “overcharge” measure of injury is flawed because it ignores what would happen to the price of the tying product in the untied world. “As a matter of elementary economic theory,” defendants argued that a seller who “has market power in the tying product and overcharges on the tied product will do two things if the tie is broken: decrease the price of the tied product and increase the price of the tying product.” Id. at 83. And since a plaintiff cannot show injury in fact if “his ‘loss’ from the defendant’s violation is offset by his ‘gain’ in the very transaction he attacks,” II Phillip E. Areeda & Herbert Hovenkamp, Antitrust Law ¶ 365c, at 245 (rev. ed.1995), the defendants contended that the tying product price had to be considered. Several authorities agree that injury must be measured by reference to the “package” price of the two products. See, e.g., Kypta v. McDonald’s Corp., 671 F.2d 1282, 1285 (11th Cir.1982) (“[Ijnjury resulting from a tie-in must be shown by establishing that payments for both the tied and tying products exceeded their combined fair market value.”); accord II Areeda & Hovenkamp, supra ¶ 383b, at 340-41 (approving this measure of injury). The issue is open in this Circuit. See In re Visa Check/MasterMoney Antitrust Litig., 192 F.R.D. at 83 n. 13.
If credit card fees would have been higher in the untied world, certification would depend (as the district court fully recognized) on which measure of injury is adopted because under the “package” measure
the fact of injury would no[t] be susceptible to class-wide proof. Specifically, merchants with relatively high amounts of credit card transactions vis a vis debit card transactions (e.g., jewelry stores) would have suffered little or no injury from the tie; indeed, they would have benefitted from it to the extent it kept credit card interchange fees down.
Id. at 83.
The district court did not decide which measure of injury applies, however, because in its view the question was obviated by a new proffer in plaintiffs’ expert’s reply declaration. According to plaintiffs new (third) conception of the untied world, defendants would have lowered off-line debit fees but “would not have raised the credit card” fees. Id. at 84 (emphasis added). The district court deemed this a “sufficient theory of class-wide injury under either the ‘package’ or the ‘tied product’ measure of injury” because under this theory even merchants with relatively high credit card volume would have been injured. Id. at 85 & n. 15. Accordingly, the *155district court saw no need to “choose between the competing line[s] of cases.” Id.
The district court’s refusal to decide how injury in fact would be measured was an error of law and thus an abuse of discretion. See Castano v. American Tobacco Co., 84 F.3d 734, 744 (5th Cir.1996) (“In order to make the findings required to certify a class action under Rule 23(b)(3) one must initially identify the substantive law issues that will control the outcome of the litigation.”). The decision was critical to the class certification motion, and involved a pure question of law; no purpose was served by a delay pending future factual development.
If the court had decided this question, and decided (correctly) that injury is measured by the package price (rather than by the price of the tied product alone), see Part II.B below, it would have encountered a subtle problem, unique to variable proportion tying cases, that is fatal to certification, see Part II.C below.
B. The Correct Measure of Injury
I would hold that the correct measure of injury in fact is the package price of the tying and tied products. The statutory text compels this rule. The injury in fact requirement is derived from § 4 of the Clayton Act, 15 U.S.C. § 15, which provides a damages cause of action only to plaintiffs who have been “injured in [their] business or property by” an antitrust violation. Id.; see J. Truett Payne Co. v. Chrysler Motors Corp., 451 U.S. 557, 562, 101 S.Ct. 1923, 68 L.Ed.2d 442 (1981); II Areeda & Hovenkamp, supra ¶ 363, at 219-27. Awarding “damages to an uninjured plaintiff is inconsistent with the statutory requirement.” II Areeda & Hoven-kamp, supra ¶ 365c, at 247. Accordingly, “the tied buyer is injured in fact only when he is forced to pay above-market prices for the sum of the tying and tied products.” X Areeda et al., supra ¶ 1769c, at 430 (citation omitted).
The price of the tying product must be taken into account because it usually does rise if the tie is broken. A rational person cannot be made to pay more for a package of two products than he would pay for the two products separately. So if a tie causes the buyer to pay more than the market price for the tied product, the buyer is most likely paying less than the price that the seller could profitably charge for the tying product if sold separately. Otherwise, the buyer would avoid the tie.5 And if somehow a plaintiff actually shows that decoupling would cause the tying product price to fall, he should be entitled to recover that much more.
Based on the text of § 4 and these economic principles, “an increasing number of courts have correctly held that the purchaser has not been injured in fact, and thus cannot recover damages, unless he shows that the combined ‘payment for both *156tying and tied products exceeds their combined fair market value.’ ” II Areeda & Hovenkamp, supra ¶ 383, at 341 (quoting Kypta v. McDonald’s Corp., 671 F.2d 1282, 1285 (11th Cir.1982)) (“A determination of the value of the tied products alone would not indicate whether the plaintiff indeed suffered any net economic harm, since a lower price might conceivably have been exacted by the [seller] for the tying product.”); see also Will v. Comprehensive Accounting Corp., 776 F.2d 665, 673 (7th Cir.1985) (“[U]nless the plaintiff shows that the package price was elevated the suit must be dismissed without further ado.”); Siegel v. Chicken Delight, 448 F.2d 43, 52 (9th Cir.1971) (“To ascertain whether an unlawful arrangement for the sale of products has caused injury to the purchaser, the cost or value of the products involved, free from the unlawful arrangement, must first be ascertained”). The competing line of cases, which measure injury by the price of the tied product alone, are, in Professor Areeda’s phrase, “quite wrong.” X Areeda, et al., supra ¶ 1769c, at 430.
C. The “Package” Measure of Injury Applied to Variable Proportion Tying Cases
If injury were measured (erroneously) by reference to the tied product alone, the interests of the class members would be aligned. Every class member processed at least some off-line debit transactions, so every class member would want the class representatives to prove that in the absence of the tie, the price of the tied product would have been as low as economic theory could support. The fact that the class members happened to process credit cards and off-line debit cards in varying proportions would be irrelevant. Indeed, as compared to the “package” measure or to a non-price theory of injury, a “claim of overcharge on the tied product is the most likely prospect for common treatment of the fact of damage.” John H. Matheson, Class Action Tying Cases: A Framework for Certification Decisions, 76 Nw. U.L.Rev. 855, 885-86 (1982) (emphasis added).
However, under the package measure described in Part II.B above, the variation among merchants of the proportion of sales using credit cards (versus off-line debit cards) has great bearing on proof of common injury. See Matheson, supra at 887; cf. Herbert Hovenkamp, Tying Arrangements and Class Actions, 36 Vand. L.Rev. 213, 249-50 (1983) (concluding that buyers subject to “variable proportion tie-in[s]” cannot prove injury in fact on a class-wide basis, and suggesting as an alternative attack the use of offensive collateral estoppel under Parklane Hosiery Co. v. Shore, 439 U.S. 322, 99 S.Ct. 645, 58 L.Ed.2d 552 (1979) (emphasis added)). To show injury, the class representatives will have to establish what both interchange fees would have been in the “but-for,” untied world — and therein lies the conflict.
Any given set of credit and debit fees that class counsel might seek to prove would have different anticipated impacts (i) on the merchant whose debit-card sales predominate, (ii) on the merchant whose credit-card and debit-card sales are in parity, (iii) on the merchant whose credit-card sales predominate-and on each merchant along the continuum. In proving injury and damages, any given economic assumption or analysis will (as the case may be) enhance, impair or preclude the antitrust claim of a merchant depending on that merchant’s location along the continuum.
Like other class members, the merchant whose debit-card sales predominate benefits by showing that the debit charge would have been much lower; but, unlike other class members, the merchant whose *157debit-card sales predominate can still achieve a substantial recovery even if the price of credit would have risen by an even greater amount. Therefore, this merchant is not compelled to make the “ ‘elusive and seldom attempted’ ” proof that “the sum of tying and tied product prices would have been lower” absent the tie. In re Visa Check/MasterMoney Antitrust Litig., 192 F.R.D. at 85 (quoting X Areeda et al., supra ¶ 1769b, at 429). (I illustrate the point with a numerical example in the margin.6)
In contrast, the merchant whose credit-card and debit-card sales are in parity would get nothing if decoupling caused credit charges to rise by as much as debit prices fell. To establish injury in fact, this merchant would be compelled to undertake the “ ‘elusive and seldom attempted’ ” proof that the sum of the untied prices would have been lower. As to the extent of injury, he would be indifferent to the relative prices of credit and debit and would do best by proving that the combination would have been as low as possible.7
The merchant whose credit-card sales predominate faces the toughest constraints. For this merchant, it is necessary but insufficient to make the “ ‘elusive *158and seldom attempted’ ” proof that the sum of the prices would have been lower. He would need to argue also that decoupling would cause the debit price to fall without causing any appreciable rise in the credit price, because even a minuscule rise in the credit price would neutralize a reduction (even a gross reduction) in the debit price. This is a tough row to hoe.8 (This is plaintiffs third theory, the one first advanced in the reply affidavit of plaintiffs’ expert.)
The three merchants at these three points on the continuum thus have different cases; their anticipated recoveries depend on and are maximized by different theories and economic assumptions; those theories and assumptions are incompatible or (at best) incoherent when advanced at once; there are winners and losers within the class; and the distribution of these competing interests creates a problem that cannot be resolved by m&class representatives even if one had a thousand of them. See Amchem Prods, v. Windsor, 521 U.S. 591, 625-26, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997) (members of a class must “possess the same interest and suffer the same injury”); General Tel. Co. v. Falcon, 457 U.S. 147, 156-57 & n. 13, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982). As far as I can see, the only unified interests served by herding these competing claims into one class are the interests served by settlement: (i) the interest of class counsel in fees, and (ii) the interest of defendants (perhaps later in the litigation) in a bundled group of all possible claimants who can be precluded by a single payment.
Certification does no service for the absent class members’ claims, viewed separately. Already, the difficulty of having to prove injury under the package measure has led plaintiffs’ counsel to propose a least-common-denominator analysis that is theoretically capable of bestowing some benefit on all class members, to the detriment of class members who would have a superior claim of recovery under other, competing analyses. Thus, plaintiffs undertake to make the “ ‘elusive and seldom attempted’ ” showing that the sum of the two fees would have been lower and that the credit card price would not have risen at all. If, as plaintiffs’ counsel evidently believes, this analysis is the only way to keep the class together, I take the point as proven that the interests of the class members are fractured and conflicted. As to the analysis itself, I have no view of its merits in this particular case. But for most absent members of the class, the analysis is unnecessary and impedes recovery under other theories that offer better prospects of recovery.
In light of the intractable conflict, I would order the district court to decertify the class. But lacking a majority on this Court for that result, I am relieved to think that the certification order is conditional and that nothing inhibits the district court from considering and reconsidering all of these issues over time, as indeed the district court is evidently disposed to do.
*159III.
Conclusion
The majority opinion exudes confidence that there is some way to bring this unmanageable and conflict-ridden litigation to an end, and that it will materialize one day in the district court. I think that is probably right, but I intuit that the only case-management tool that will bring about that end will be settlement, and that it will be coerced by abuses that Rule 23(f) was specifically designed to correct.
. Many district courts have ignored Judge Friendly's admonition in Abrams. But as the trial judge in this case recognized, those courts “state the rule that individual damages questions will not bar class certification In conclusory fashion without assessing whether individual questions would predominate if each damages action had to be tried separately.” In re Visa Check/MasterMoney Antitrust Litig., 192 F.R.D. at 86 n. 19 (emphasis added); cf. Szabo v. Bridgeport Machines, 249 F.3d 672, 675 (7th Cir.2001) (granting Rule 23(1) review because "the district court’s decision to certify the class implie[d] that important legal principles have evaded attention by appellate courts. At critical junctures the district judge cited only decisions by other district judges, most in cases later settled and thus not subject to appellate consideration.”).
. The right to assert the defense against each and every class member is guaranteed by the Rules Enabling Act. See Eisen v. Carlisle & Jacquelin, 479 F.2d 1005, 1014 (2d Cir.1973) (recognizing that under the Act, Rule 23 cannot “ 'abridge, enlarge or modify any substantive right’ ”) (quoting 28 U.S.C. § 2072), vacated on other grounds, 417 U.S. 156, 94 S.Ct. 2140, 40 L.Ed.2d 732 (1974).
. I agree with the majority that the "honor all cards” rule precluded any merchant from mitigating his alleged losses down to zero, Majority Op. at 17, but the issue that challenges manageability is the effect of channeling on the amount of damages. See in re Visa Check / Mastermoney Antitrust Litig., 192 F.R.D. at 86 ("The fact of injury is binary; it is either there or it is not. The extent of injury is another question.”).
. I also assume arguendo that plaintiffs can use common proof to show that the tie had an anticompetitive effect on the relevant market, a showing required of all private antitrust enforcers regardless of whether they pursue a "per se” or rule of reason claim. See Balaklaw v. Lovell, 14 F.3d 793, 801 & n. 17 (2d Cir.1994).
. The Areeda treatise offers the following example:
Suppose a defendant's profit maximizing price for product A alone is $100 and that the prevailing market price for product B is $20. If the defendant’s version of product B is worth $ 1 less to the plaintiff than rival brands (or if a tie is otherwise burdensome in the amount of $1), the defendant tying the two together cannot profitably charge more than $119 for the package. Phrased another way: if he charges $25 for tied product A, he must reduce the price of tying product A to $94.
II Areeda & Hovenkamp, supra ¶ 365 at 246 n. 48 (emphasis added). Thus, if the tie is broken, the price of the tying product will increase back to $100.
This example describes a fixed proportion tie, but even in the variable proportion context "tied-product premiums are often offset by tying-product discounts below the price that would have prevailed without tying.” X Areeda et al., supra ¶ 1769 at 431.
. Suppose that the interchange fees in the tied world (the actual world) are $10 per $100 purchase for both credit cards and offline debit cards (in fact, the percentages are much smaller but the $10 figure is useful for illustrative purposes). Thus, regardless of an individual merchant's mix of credit and debit, for every 100 transactions he pays 100 x $10 or $1000.
To establish injury in fact under the package measure, any merchant would have to show that in the untied world, he would have paid less than $1,000. As noted in the text, this will require establishing what both interchange fees would have been in the untied world, i.e., the fee on credit cards would have been "$C” and the fee on off-line debit cards would have been "$D”.
To assess the effect of different price combinations on the merchant "whose debit-card sales predominate,” let us say that for every 100 transactions, he processes 30 by credit and 70 by off-line debit. In showing the fact of injury, this merchant could select (if provable) combinations of "$C” and "$D” that would sum over $20, thus obviating the need to make " 'elusive and seldom attempted' " showing that "the sum of tying and tied product prices would have been lower” absent the tie. In re Visa Check/MasterMoney Antitrust Litig., 192 F.R.D. at 85 (quoting X Areeda et al., supra ¶ 1769b, at 429). For example, if C = $19 and D = $5, then C + D would be $24 but this merchant would have been injured in fact by $150 (30 x $19 + 70 x $4 = $570 + $280 = $850).
In terms of the extent of his injury, the 30%/70% merchant would greatly prefer a theory positing that C would have been $16 and D would have been $3 to a theory that the prices would have been $10 and $9, even though in both cases the sum would be $19. He could recover $310 under the former version of events (30 x $16 + 70 x $3 = $690), but only $70 under the latter (30 x $10 + 70 x $9 = $930).
. By plugging in the hypothetical prices of credit and debit discussed in the previous footnote, it is easy to see how the interests of this merchant differ from those of the merchant whose debit sales predominate. For every 100 transactions, this merchant processes 50 credit sales and 50 debit sales. If C would have been $19 and D $4, this merchant would actually have paid more in the untied world then he did under the tie (50 x $19 + 50 x $4 = $1,150). In the words of the district court, he "would have benefitted from [the tie] to the extent it kept credit card interchange fees down.” In re Visa Check/Master-Money Antitrust Litig., 192 F.R.D. at 83.
And while the respective prices of credit and debit mattered a great deal to 30%/70% merchant in showing the extent of injury, they would not matter at all to the 50%/50% merchant. The latter would recover $50 by advancing an economic theory that the prices would have been $16 and $3 (50 x $16 + 50 x $3 = $950) or $10 and $ 9 (50 x $10 + 50 x $9 = $950). The 50%/50% merchant could make a disinterested determination of which of the two theories would be easier to prove.
. Assume that this merchant processed 70 credit transactions for every 30 debit sales. If decoupling would have resulted in prices of $19 and $4, then this merchant would have been hurt by a whopping $450 (70 X $19 + 30 x $4 = $1,450). He also would have been hurt if the prices had been $16 and $3 (70 x $16 + 30 X $3 = $1,210) or even $16 and $0 because 70 x $16 is more than $1000. On the other hand, if the prices were $10 and $9, he could recover $30 (70 x $10 + 30 x $9 = $970), placing his preferences directly at odds with the preferences of the 30/70 merchant (who preferred the $16 and $3 prices to the $10 and $9). By extension, a merchant whose ratio of credit to off-line debit transactions was extremely high would not be able to recover under any theory positing a credit card fee of over $10 unless the debit card fee became virtually a free good.